[Feature] Spec-Overlap supporting DP-ATTN; PD-Disaggregation; npugraph mode (#12443)

test/srt/ascend/test_ascend_deepseek_mtp.py

@@ -43,6 +43,9 @@ class TestAscendDeepSeekMTP(CustomTestCase):
             32768,
             "--tp-size",
             16,
+            "--dp-size",
+            2,
+            "--enable-dp-attention",
             "--speculative-algorithm",
             "NEXTN",
             "--speculative-num-steps",
@@ -55,6 +58,8 @@ class TestAscendDeepSeekMTP(CustomTestCase):
 
         cls.extra_envs = {
             "SGLANG_NPU_USE_MLAPO": "1",
+            "SGLANG_ENABLE_SPEC_V2": "1",
+            "SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
         }
         os.environ.update(cls.extra_envs)
 
@@ -91,26 +96,6 @@ class TestAscendDeepSeekMTP(CustomTestCase):
                 finally:
                     kill_process_tree(process.pid)
 
-    def test_b_throughput(self):
-        for model in self.models:
-            with self.subTest(model=model):
-                print(f"##=== Testing throughput: {model} ===##")
-
-                output_throughput = run_bench_offline_throughput(
-                    model,
-                    [
-                        *self.common_args,
-                    ],
-                )
-
-                print(f"##=== {model} throughput: {output_throughput} ===##")
-
-                if is_in_ci():
-                    self.assertGreater(
-                        output_throughput,
-                        TEST_MODEL_MATRIX[model]["output_throughput"],
-                    )
-
 
 if __name__ == "__main__":
     unittest.main()